1 October 2003
A serious bug in the application of 1-D filters with gaps may cause
Sherpa to give incorrect fit results. This bug is present in CIAO 3.0
and 3.0.1.
DETAILS:
The algorithm that constructs the grid upon which models are evaluated
does not properly take into account filter gaps. If a user defines a 1-D
filter with gaps in it, then the calculation of the model predicted
values produces many more counts than what it is correct
in a bin at the edge of each gap.
Sherpa will NOT issue an error, but the results of fitting with this
filter will be INCORRECT.
NOTE that this bug applies *only* to 1-D filters. This bug DOES NOT
apply to 1-D continuous filters or any 2-D filters.
EXAMPLE:
in the following filter, all the counts in the excluded region (7-8 Å)
will be included in the last bin of the 6-7 Å region:
sherpa> ignore allsets all
sherpa> notice 1 wave 6:7
sherpa> notice 1 wave 8:9
A typical instance of such a filter could be when trying to fit a
continuum excluding specific emission/absorption lines.
WORKAROUND: None.
The only way to avoid getting silently erroneous results is to avoid
gaps in filters.
FIX: Will be available in a future CIAO patch
This archive was generated by hypermail 2b29 : Wed May 15 2013 - 01:00:12 EDT